init commit

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+vit.pth filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,69 @@
+import gradio as gr
+import os
+import torch
+
+from model import create_vit
+from timeit import default_timer as timer
+from typing import Tuple, Dict
+
+# Setup class names
+with open("class_names.txt", "r") as f:
+  class_names = [name.strip() for name in f.readlines()]
+
+### Model and transforms preparation ###
+# Create model and transforms
+model, _, _, transforms = create_vit(output_shape=101, classes=class_names)
+
+model = torch.compile(model)
+
+# Load saved weights
+model.load_state_dict(
+    torch.load(f="vit.pth",
+               map_location=torch.device("cpu")) # load to CPU
+)
+
+### Predict function ###
+def predict(img) -> Tuple[Dict, float]:
+  # Start a timer
+  start_time = timer()
+
+  # Transform the input image for use with the model
+  img = transforms(img).unsqueeze(0) # unsqueeze = add batch dimension on 0th index
+
+  # Put model into eval mode, make prediction
+  model.eval()
+  with torch.inference_mode():
+    # Pass transformed image through the model and turn the prediction logits into probaiblities
+    pred_probs = torch.softmax(model(img), dim=1)
+
+  # Create a prediction label and prediction probability dictionary
+  pred_labels_and_probs = {class_names[i]: float(pred_probs[0][i]) for i in range(len(class_names))}
+
+  # Calculate pred time
+  end_time = timer()
+  pred_time = round(end_time - start_time, 4)
+
+  # Return pred dict and pred time
+  return pred_labels_and_probs, pred_time
+
+### 4. Gradio app ###
+# Create title, description and article
+title = "A ViT cifar10 Classifier"
+description = "An [ViT feature extractor](https://huggingface.co/google/vit-base-patch16-224) computer vision model to classify images on the [10 classes of the cifar10 dataset](https://huggingface.co/datasets/cifar10). [Source Code Found Here](https://colab.research.google.com/drive/1j4NbiMpCqmXN1xw9e2_r77gMdr3WpMnO?usp=drive_link)"
+article = "Built with [Gradio](https://github.com/gradio-app/gradio) and [PyTorch](https://pytorch.org/). [Source Code Found Here](https://colab.research.google.com/drive/1j4NbiMpCqmXN1xw9e2_r77gMdr3WpMnO?usp=drive_link)"
+
+# Create example list
+example_list = [["examples/" + example] for example in os.listdir("examples")]
+
+# Create the Gradio demo
+demo = gr.Interface(fn=predict, # maps inputs to outputs
+                    inputs=gr.Image(type="pil"),
+                    outputs=[gr.Label(num_top_classes=5, label="Predictions"),
+                             gr.Number(label="Prediction time (s)")],
+                    examples=example_list,
+                    title=title,
+                    description=description,
+                    article=article)
+
+# Launch the demo
+demo.launch()

class_names.txt

@@ -0,0 +1,91 @@
+airplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truckairplane
+automobile
+bird
+cat
+deer
+dog
+frog
+horse
+ship
+truck

model.py

@@ -0,0 +1,60 @@
+import torch
+import torchvision
+from torch import nn
+from torchvision import transforms
+from transformers import ViTForImageClassification
+from transformers import ViTImageProcessor
+from typing import List
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+def create_vit(output_shape:int, classes:List, device:torch.device=device):
+  """Creates a HuggingFace ViT model google/vit-base-patch16-224
+
+  Args:
+    output_shape: The output shape
+    classes: A list of classes
+    device: A torch.device
+
+  Returns:
+    A tuple of the model, train_transforms, val_transforms, test_transforms
+  """
+  id2label = {id:label for id, label in enumerate(classes)}
+  label2id = {label:id for id,label in id2label.items()}
+
+  model = ViTForImageClassification.from_pretrained('google/vit-base-patch16-224',
+                                                  num_labels=len(classes),
+                                                  id2label=id2label,
+                                                  label2id=label2id,
+                                                  ignore_mismatched_sizes=True)
+
+  for param in model.parameters():
+    param.requires_grad = False
+
+  # Can add dropout here if needed
+  model.classifier = nn.Linear(in_features=768, out_features=output_shape)
+
+  #https://github.com/NielsRogge/Transformers-Tutorials/blob/master/VisionTransformer/Fine_tuning_the_Vision_Transformer_on_CIFAR_10_with_PyTorch_Lightning.ipynb
+  processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224")
+  image_mean = processor.image_mean
+  image_std = processor.image_std
+  size = processor.size["height"]
+
+  normalize = transforms.Normalize(mean=image_mean, std=image_std)
+  train_transforms = transforms.Compose([
+      #transforms.RandomResizedCrop(size),
+      transforms.Resize(size),
+      transforms.CenterCrop(size),
+      transforms.RandomHorizontalFlip(),
+      transforms.ToTensor(),
+      normalize])
+
+  val_transforms = transforms.Compose([
+      transforms.Resize(size),
+      transforms.CenterCrop(size),
+      transforms.ToTensor(),
+      normalize])
+
+  test_transforms = val_transforms
+
+  return model.to(device), train_transforms, val_transforms, test_transforms

requirements.txt

@@ -0,0 +1,4 @@
+torch==2.1.0
+torchvision==0.16.0
+gradio==3.50.2
+transformers==4.35.0

vit.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f92179e0f94cb25af399d2e2e324a2390fd9e9c842728f866451b6b3d7db625
+size 343297010